Reading AI-generated communications is enslavement by the robots
Humans have finite time on this earth. It’s an insult and ethical crime to cause someone to contemplate AI content under false pretenses.
Google’s Gemini AI ad for the Olympics—in which a father “helps” his small daughter produce a fan letter to her favorite Olympian by simply asking the LLM to write one—has come in for well deserved derision, most delightfully from the Washington Post’s Alexandra Petri. Her angle on the problem, which is that AI does not know your thoughts and feelings, and thus its transmissions are vacuous and generic, is the right way to think about what writing means for humans. This point has been made elsewhere as well, and remains a philosophical challenge with AI-driven writing.
The flip side of this, though, gets less worry, but it is to my view the bigger problem: What about the reader of an AI-generated letter? The Olympian hurdler in this case, but it’s also true of your boss who gets a cooked up status letter; your podcast listener who sits through an auto-generated script; the aspiring romantic sifting through a computer-drafted dating profile; or your teacher who marks up a blast of LLM homework with a red pen.
If the audience of your work knew that the material they were spending their time reading, or hearing, or watching, came not from you but from some computer who pretended to be a person for six quick seconds, would they still think it valuable to spend their time digesting and considering what you give them?
We humans are here on this earth for the briefest of moments, within an infinity of darkness. Our time is valuable, in the most existential sense. Humans have always chosen to read books, correspond with each other, watch films, and listen to preachers and storytellers because human communication is, as Petri has it, “a method for taking thoughts and images and stories out of your brain and putting them into someone else’s brain.” We choose to pay attention because we want to know and understand the experiences of other humans. We have trusted in those transmissions because we understand that crafting anything from a novel all the way down to a quick employee status memo requires the brainpower of a human.
But what happens when human brainpower is no longer involved? Since the earth-shattering advent of ChatGPT in late 2022, the internet has already become awash in AI-generated gloop, generic but reasonable stuff that masquerades as human work. That’s bad enough.
But if you send me a letter, or an email, or essay, or a memo, and you imply to me that it comes from you—a person—but it actually came from a computer, then you’re defrauding me of my limited time on earth. The five minutes, or ten minutes, or hour that I spend trying to form a picture of your mind, in my own mind, is time thrown away on a con. The person who has vouched for the stuff is the conman, but worse, they are a traitor—a handmaiden of the robots.
The computers have infinite resources. The computers can produce more legible text, faster, than every literate human could possibly read, non-stop, in real time. All of humanity could be kept docile by the outputs of the LLMs, and their capacity seems to be only increasing. It costs the robot only some kilowatts and a second or two to produce that faux fan letter. But it costs the Olympian something real to read it, think about it, and possibly even reply to it. Aza Raskin and Tristan Harris talk about how “language is the operating system of human culture” and go on to observe that AI’s new language ability allows it to “hack and manipulate the operating system of civilization.” Please do the rest of your fellow humans a favor: don’t drag the Trojan Horse through our gates just because you don’t feel like doing the status memo today. (At the very least do your team the courtesy of cutting out the middleman and just send them your prompt text.)
LLMs seem to have plenty of genuinely valuable uses—as interlocutors for question-driven learning, as summarizers of complex material, even (yes) as helpful editors of human writing. But it’s notable that Google allowed this advertisement to reach audiences without realizing how strange the implications of their specific example were. It suggests that the groups developing AI technology may not spend as much time as they should considering the value of their fellow humans’ time and humanity.
This article was not written by a large language model. But how would you feel if this note said that it was?
